多项式曲面拟合和BP神经网络GPS高程拟合方法的比较研究

为了研究不同GPS高程拟合方法的拟合精度及差异性,选择二次多项式曲面拟合法和BP神经网络法进行比较研究.对已知数据进行分组,采用二次多项式曲面拟合法和BP神经网络法分别对各组数据进行处理、分析,计算精度指标,比较拟合点的分布、拟合点的数量和拟合方法对精度的影响.试验算例表明采用相同方法时拟合点的数量越多、分布越均匀拟合精度越高;在点的数量相同且较多,分布都相对均匀的前提下,BP神经网络法拟合的精度高于二次多项式曲面拟合法;在点的数量相同且较少,分布都相对均匀的前提下,BP神经网络法拟合的精度低于二次多项式曲面拟合法.

A comparative study of GPS elevation fitting method： polynomial surface fitting method and BP neural network method

In order to research the accuracy and diversity of different GPS elevation fitting methods,two methods of quadratic polynomial surface fitting and BP neural network are provided.The known data are categorized,and the two methods are used to process,analyze and compute the standards of accuracy,the effect on accuracy from the distribution of fitted points,the number of fitted points and the elevation fitting method.The results demonstrate in the experimental example：with the same fitting method,the more the fitted points are distributed,the higher precision of fitting is.On the condition of using the same less number of fitting points with balanced distribution,the precision of fitting by the quadratic polynomial surface fitting method is higher than by the BP neural network method;on the condition of using the same more number of fitting points,the result is opposite.